Flux.jl/examples/MNIST.jl

using Flux, MNIST
using Flux: accuracy

data = [(trainfeatures(i), onehot(trainlabel(i), 0:9)) for i = 1:60_000]
train = data[1:50_000]
test = data[50_001:60_000]

m = @Chain(
  Input(784),
  Affine(128), relu,
  Affine( 64), relu,
  Affine( 10), softmax)

# Convert to MXNet
model = mxnet(m)

# An example prediction pre-training
model(tobatch(data[1][1]))

Flux.train!(model, train, η = 1e-3,
            cb = [()->@show accuracy(m, test)])

# An example prediction post-training
model(tobatch(data[1][1]))
a bunch of stuff 2016-08-25 21:49:21 +00:00			`using Flux, MNIST`
use fancy new callback api 2017-05-01 13:23:48 +00:00			`using Flux: accuracy`
init 2016-04-01 21:11:42 +00:00
update mnist example 2017-02-02 04:39:41 +00:00			`data = [(trainfeatures(i), onehot(trainlabel(i), 0:9)) for i = 1:60_000]`
sort-of working mnist example 2016-09-29 20:28:53 +00:00			`train = data[1:50_000]`
			`test = data[50_001:60_000]`
init 2016-04-01 21:11:42 +00:00
replace old shape inference system 2017-03-17 16:34:51 +00:00			`m = @Chain(`
init 2016-04-01 21:11:42 +00:00			`Input(784),`
dense -> affine 2016-11-14 22:16:00 +00:00			`Affine(128), relu,`
			`Affine( 64), relu,`
			`Affine( 10), softmax)`
init 2016-04-01 21:11:42 +00:00
update logreg example 2017-03-09 00:23:04 +00:00			`# Convert to MXNet`
			`model = mxnet(m)`
run with what we have 2016-08-25 16:26:52 +00:00
example predictions 2016-10-04 20:11:03 +00:00			`# An example prediction pre-training`
use fancy new callback api 2017-05-01 13:23:48 +00:00			`model(tobatch(data[1][1]))`
example predictions 2016-10-04 20:11:03 +00:00
use fancy new callback api 2017-05-01 13:23:48 +00:00			`Flux.train!(model, train, η = 1e-3,`
			`cb = [()->@show accuracy(m, test)])`
example predictions 2016-10-04 20:11:03 +00:00
			`# An example prediction post-training`
use fancy new callback api 2017-05-01 13:23:48 +00:00			`model(tobatch(data[1][1]))`